Research On Intrusion Detection Based On Deep Learning

In recent years,information technology is becoming more and more important in people's life,and the connection between the two is also increasing,the network brings people all kinds of convenience,at the same time,people are facing more and more diverse network security problems:from stealing the password of system administrator,stealing private information and privacy,to attacking the service and information system of schools,governments,hospitals and other units,destroying the confidentiality of the system,integrity and availability,resulting in huge social and economic losses.Therefore,the research of network intrusion detection technology is more and more necessary.Based on the current application of deep learning,deep learning algorithm plays an irreplaceable role in different fields,and also develops rapidly in the field of intrusion detection.Compared with the traditional intrusion detection algorithm,the intrusion detection model based on deep learning algorithm has the advantages of higher accuracy and more intelligence.In the first mock exam,the integration learning theorysolves the defect of single model.In this paper,deep and ensemble learning algorithm is used to improve the accuracy and ability of intrusion detection.The main work includes:(1)This paper introduces the research background and significance of intrusion detection,expounds the current research trend and research status at home and abroad.(2)This paper summarizes the basic intrusion detection technology and the theoretical knowledge of machine learning algorithms.(3)Due to the imbalance of network intrusion data samples,the detection ability and accuracy of the algorithm are weak,and the representation ability of traditional machine learning algorithm for complex functions is limited.Bigru is used as the core of the model,focal loss function is used to adjust the weight of minority samples,nsl-kdd data set is used for experiment,and Adam optimization algorithm is used to update the weight matrix continuously,so as to achieve the best effect of the model.The experimental results show that the detection ability of this model is stronger than that of traditional machine learning.(4)KSLSMOTE sampling algorithm is proposed.The algorithm improves the problem of smote which is easy to generate noise samples and the way of generating new samples.It breaks through the limitation of single linear interpolation.Finally,KFCM is used to cluster and undersampling most samples,so as to improve the unbalanced data set.(5)On the basis of the KSLSMOTE algorithm,the first mock exam is limited and the accuracy is low.Using ensemble learning ensemble classifier.According to the data characteristics of multi-classifier after full learning and its advantages,an experiment is designed and compared with other algorithms.The innovations of this paper include:(1)A network intrusion detection method based on bigru is proposed.Bigru neural network is used to learn and fit the characteristics of data samples.Meanwhile,for the imbalance of data samples in intrusion data set,focal loss subtraction is used to improve the data imbalance,so as to improve the learning ability of the model for minority samples and solve the low accuracy of minority samples.Batch normalization mechanism and dropout algorithm were normalized the data data to improve the generalization ability.The nsl-kdd data set is used for experiment,and the experimental results show that the accuracy and recall rate of the method.(2)KSLSMOTE sampling algorithm is proposed.This algorithm not only improves the method of generating samples in SMOTE algorithm,but also avoids the noise of generating new samples during SMOTE algorithm.and integrates KFCM algorithm: firstly,the k-nearest neighbors of a few sample points are analyzed to determine whether to generate new samples or the direction of generating new samples.At the same time,the values of each dimension of the samples are used to generate new samples,which improves the noise of new samples generated by smote Defects.Then KFCM algorithm is used to cluster the data set,and most of the samples are undersampled to balance the data set.On the basis of KSLSMOTE algorithm,a new hierarchical integration model is constructed by using stacking integration technology.SVM,LR,KNN and VAE are used as the base classifiers of the model,which combines the advantages of each base classifier and enhances the detection ability of the model.